Getting good returns on AI spending is the hottest enterprise topic of the day, and AI consultants have a neat cost-saving hack.<\/p>\n
Business Insider has reported extensively about how users should route their tasks to the right AI models<\/a> to get the best bang for their buck. But AI consultants<\/a> are going one step further in the AI efficiency discussion, saying that the most expensive frontier models should be treated like the expensive consultants and advisors they are.<\/p>\n “You wouldn’t use the most expensive lawyer for filing some routine things that less expensive lawyers can,” said Ameya Kanitkar, the CTO of the San Francisco-based AI measurement platform Larridin.<\/p>\n Speaking to Business Insider in July, Kanitkar said models like Fable 5, the most powerful and token-burning model from Anthropic, should be used to create a road map of workflows, while cheaper and smaller models should execute them.<\/p>\n “The advisory model basically plans things, breaks down the problems into smaller sets, and has the complete context of how everything’s going to work,” he said. “And then sub-tasks are delegated to cheaper models.”<\/p>\n Larridin advises companies on their AI use, giving them visibility into how effectively their employees are using AI tools and how to achieve greater ROI from their AI spend.<\/p>\n Lightweight models should do the heavy-lifting<\/p>\n In the same spirit, Michael Murphy, a partner at the Sydney-based AI transformation consultancy Adaptovate, said it’s not a great use of company dollars to send “the most powerful model out there to do something that’s replacing a Google check or transcribing meeting minutes or creating a creative brief.”<\/p>\n Murphy said frontier models should be used to strategize, create the initial builds of a new app or website, or tasks that require complex thinking. Then, companies should figure out which “lightweight flashlight models” are most appropriate for day-to-day tasks, he said.<\/p>\n It’s not just consultants who are vouching for this tactic \u2014 industry leaders have also talked about the idea of using frontier models as advisors.<\/p>\n In an X post in June, Coinbase CEO Brian Armstrong<\/a> said he anticipated that “80% of workloads will be running on 99% cheaper models within 12-18 months.”<\/p>\n The best models should be kept for “IQ maxxing,” he said, such as scientific breakthroughs or agent orchestration.<\/p>\n AI ROI is a hot priority for executives<\/p>\n Cost-saving tactics like these are becoming more popular as companies become increasingly concerned about not getting\u00a0proportional returns<\/a>\u00a0on their AI spending.<\/p>\n Many have abandoned tokenmaxxing, a trend in which companies gave their employees free rein to experiment with AI, urging them to burn as many tokens as possible. Some, like Duolingo, even made AI usage a performance metric.<\/p>\n Now, companies are being more conservative, thinking about how to get the most value out of their spend.<\/p>\n Several budget hacks have emerged, such as model routing or using open-source Chinese models like Moonshot AI’s Kimi K3<\/a> or Z.ai’s GLM-5.2.<\/p>\n A new wave of startups is cashing in on AI ROI, becoming a C-suite priority. The startups, known as AI-routing companies that help steer developers toward different AI models and monitor for overspending, are becoming investor favorites.<\/p>\n